AI-first company

Claim check

“Employment of young workers in AI-exposed jobs fell 13% (Stanford)”

The 13% is the headline of the first version of the paper, from August 2025. It is a gap between 22–25-year-olds in the occupations most and least exposed to AI, not a fall in employment. In levels, their employment in the most exposed occupations fell 6%. The paper has been revised twice. The November 2025 version gave 16%, and the August 2026 version gives 19% on a simpler measure. With current data and the first version’s filters, the estimate behind the 13% can no longer be told apart from chance for the most exposed occupations.

By
Last reviewed
Corrections
Report an error
Verdict Needs correcting

The source exists, but the claim gets the year, the population or the wording wrong.

Kind of figure Measured

Counted from records or measured in an experiment.

Holds for 22–25-year-olds in US payroll records from ADP, compared by how exposed their occupation is to AI. A relative gap that describes a pattern; the authors do not claim AI caused it. Cite the version and its date: 13% (August 2025), 16% (November 2025), 19% on a different measure (August 2026).

The claim

Employment of young workers in AI-exposed jobs fell 13%, according to a Stanford study by Brynjolfsson, Chandar and Chen.

How the claim travelled

Each step is a publication that passed the claim on, oldest first, with its exact words and what changed.

  1. Research

    Stanford Digital Economy Lab (Brynjolfsson, Chandar and Chen) · archived copy

    “We find that since the widespread adoption of generative AI, early-career workers (ages 22-25) in the most AI-exposed occupations have experienced a 13 percent relative decline in employment even after controlling for firm-level shocks.”

    What changed: Origin. The authors compared employment in ADP’s US payroll records by age and by how exposed each occupation is to AI, from late 2022 to July 2025. The 13% is a gap between the most and least exposed occupations, after removing changes that hit a whole firm. In levels, employment of 22–25-year-olds in the most exposed occupations fell 6%.

  2. Press

    LeadDev (Chantal Kapani) · archived copy

    “Workers aged 22 to 25 are the most exposed to artificial intelligence, suffering a decline in employment of 13%, according to a new Stanford University study. … Overall U.S. employment has continued to grow since late 2022, yet employment among young workers in AI-exposed roles such as software development, customer service, and data analysis fell by 13% through July 2025.”

    What changed: “Relative” is dropped, and the gap becomes a fall: “fell by 13%”. The age group, not the occupations, becomes “the most exposed to artificial intelligence”. Other outlets that week differed: AfroTech kept “relative”, and Sinclair’s syndicated “Fact Check Team” wrote “a 13 percent drop”.

  3. Research

    Stanford Digital Economy Lab (Brynjolfsson, Chandar and Chen) · archived copy

    “Early-career workers (ages 22-25) in AI-exposed occupations experienced 16% relative employment declines, controlling for firm-level shocks, while employment for experienced workers remained stable.”

    What changed: The revised paper adds data up to September 2025. The same kind of firm-adjusted gap is now 16%. From this version on, the 13% is out of date.

  4. Research

    Federal Reserve Bank of Dallas (Tyler Atkinson and Shane Yamco) · archived copy

    “Workers age 22 to 25 in the most AI-exposed occupations have experienced a 13 percent decline in employment since 2022, a recent study by researchers at Stanford University found.”

    What changed: Two months after the 16% version, a central bank’s research note repeats the first version’s 13%. “Relative” is dropped, so the gap reads as a fall in employment. The note’s own survey data, for 20–24-year-olds, show a much smaller decline.

  5. Blog

    AI Weekly (Alexis Dufresne) · archived copy

    “Early-career workers ages 22 to 25 in the most AI-exposed occupations show roughly a 13 percent relative decline in employment since ChatGPT's 2022 launch.”

    What changed: A newsletter summarizing a Stanford policy brief gives the first version’s 13%, eight months after it was replaced. The brief, from the Stanford Institute for Economic Policy Research, gives no percentage for young workers and cites the November 2025 version. “Roughly” is added.

  6. Research

    Stanford Digital Economy Lab (Brynjolfsson, Chandar and Chen) · archived copy

    “(2) However, employment of young workers (ages 22–25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap.”

    What changed: Follow-up. The current version, with data up to June 2026, headlines a simpler measure: 19% below where employment would be had it kept pace with less exposed occupations. On that measure the gap was 15% with data up to July 2025. With the first version’s filters, the firm-adjusted estimate for the most exposed occupations is no longer clearly different from zero.

What the primary source says

The paper is Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen of Stanford University. It is a working paper of the Stanford Digital Economy Lab, which paid for the work, and it has not been peer-reviewed. ADP, the largest payroll software company in the United States, supplied the data. ADP also funds the lab as a member of its Corporate Affiliate Program. Erik Brynjolfsson also co-founded Workhelix, a company that analyses AI adoption in firms, and co-chairs its board. Three versions have circulated: 26 August 2025, 13 November 2025 and August 2026 (announced on 12 August 2026).

How it was done. The authors used ADP’s monthly payroll records for millions of US workers. They sorted occupations into five equal groups by exposure to AI: how many of an occupation’s tasks AI could do. Exposure is not a measure of whether firms use AI. They then followed employment by age group from late 2022, when ChatGPT was released.

What it found. In the current version, with data up to June 2026, employment of 22–25-year-olds in the two most exposed groups of occupations fell about 11%. In the three least exposed groups, employment of the same age group grew about 10%. Experienced workers show no such gap. The gap comes mainly from fewer young people being hired, not from more people leaving. The authors find “no evidence of widespread, economy-wide job displacement”. They call their results “early, descriptive indicators—canaries in the coal mine—rather than causal estimates”.

Where the 13% appears. In the abstract of the first version: “early-career workers (ages 22-25) in the most AI-exposed occupations have experienced a 13 percent relative decline in employment even after controlling for firm-level shocks.” The authors calculated it by comparing 22–25-year-olds in the most exposed fifth of occupations with those in the least exposed fifth, after removing changes that hit everyone at the same firm. The text gives it as “a 12 log-point decline in relative employment”. In levels, the same version says, employment of 22–25-year-olds in the most exposed occupations fell 6% from late 2022 to July 2025. The November 2025 version gave 16% on the same kind of measure. The August 2026 version moved to a simpler measure, the 19% shortfall, which “requires no modeling choices”. On that measure the gap was 15% with data up to July 2025.

What is wrong with the claim

  • The 13% is a gap, not a fall. The first version reports a “relative decline”: young workers in the most exposed occupations against young workers in the least exposed. Their employment fell 6% in levels over the same period, while employment in less exposed occupations grew.
  • It quotes the first of three versions. The November 2025 version gave 16%. The current version, of August 2026, gives 19% on a different measure. Repeats in 2026, including a research note by the Federal Reserve Bank of Dallas, still give 13%.
  • The estimate behind the 13% has weakened. With the current data and the first version’s filters, the estimate for the most exposed fifth of occupations falls to −5.3 log points and is “no longer statistically significant”: it can no longer be told apart from chance. Under the filters the authors use now, it is −6.8 log points, about half the first version’s estimate, and still statistically significant. The estimate for the next most exposed fifth holds at −12.8 log points. The authors write that these firm-adjusted results “were more sensitive to changes we made in the data pipeline”.
  • It does not show that AI caused the gap. The authors call their results descriptive. The current version reports that the gap shrinks when workers’ education is taken into account, that some of it started before generative AI, and that it is larger in ADP’s records than in national surveys.
  • “Young workers” means 22–25-year-olds in ADP’s US payroll records. The claim drops the age band and the country. The authors write that “the magnitude of the divergence appears specific to the ADP sample, but its direction is consistent.”

What other sources found

A working paper by Hosseini and Lichtinger used résumé data on 65 million US workers. At firms that hired staff to build generative AI into their work, junior employment fell about 9% relative to other firms within six quarters, mainly because those firms hired fewer juniors. That is a different measure: firms that adopted AI, not occupations exposed to it.

The Federal Reserve Bank of Dallas note in the lineage above also looked at the US Current Population Survey, a monthly household survey run by the US government. It found “some decline in employment for young workers in the most AI-exposed occupations”: the share of all 20–24-year-olds employed in the most exposed third of occupations “slips from 16.4 percent in November 2022 … to 15.5 percent in September 2025”. It found that layoffs do not explain the decline. Across sources, the direction agrees: fewer young people are being hired into AI-exposed work in the United States. The size differs a lot by data set.

The paper is worth citing for the direction and the source of the gap: fewer young people hired into occupations where AI can do much of the work. Cite the version and its date, and cite the gap as a gap, not as a fall in employment.

What to cite instead

In US payroll records from ADP up to June 2026, employment of 22–25-year-olds in the occupations most exposed to AI was 19% below where it would have been had it kept pace with less exposed occupations, mainly because fewer young people were hired; experienced workers showed no such gap (Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab working paper, revised 12 August 2026). The authors describe these as “early, descriptive indicators” and not proof that AI caused the gap.

Sources

  1. Brynjolfsson, Chandar and Chen (26 August 2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab working paper, first version. The version the claim quotes; the original address now leads to the current version, so this is an archived copy. web.archive.org
  2. Brynjolfsson, Chandar and Chen (13 November 2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Second version (archived copy). web.archive.org
  3. Brynjolfsson, Chandar and Chen (August 2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Current version, with Online Appendix Section L on the changes between versions. digitaleconomy.stanford.edu
  4. Stanford Digital Economy Lab. Canaries in the Coal Mine? Publication page, with the authors’ note of 12 August 2026. digitaleconomy.stanford.edu
  5. Kapani (27 August 2025). Young devs see steepest job losses in AI-exposed roles, Stanford study finds. LeadDev. leaddev.com
  6. Atkinson and Yamco (6 January 2026). Young workers’ employment drops in occupations with high AI exposure. Federal Reserve Bank of Dallas. dallasfed.org
  7. Stanford Institute for Economic Policy Research (2026). What is really happening to jobs? Separating AI hype from reality. Policy brief. siepr.stanford.edu
  8. Dufresne (26 July 2026, updated 11 August 2026). Stanford SIEPR: AI Hits New Grads, Not Aggregate Jobs Yet. AI Weekly. aiweekly.co
  9. Hosseini and Lichtinger (2025, version of 6 May 2026). Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data. SSRN working paper 5425555. papers.ssrn.com
  10. Workhelix (2026). About Us. Company page that lists Erik Brynjolfsson as co-founder and co-chairman of the board. workhelix.com
  11. Brynjolfsson, Chandar and Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab working paper (revised 12 August 2026). What this study did and found
  12. Hosseini and Lichtinger (2025). Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data. SSRN working paper 5425555 (version of 6 May 2026). What this study did and found